Research On Detecting Non-aligned Double JPEG Compression For Color Image

In recent years,with the rapid development of multimedia technology,various powerful and easy-to-use image processing software is widely used,it has become very common for images to be edited and tampered with.JPEG compression is the popular image compression standard,a large number of images will be stored in JPEG format.In reality,tampers mostly use image editing software to manipulate images and then save them in JPEG format,so nonaligned double JPEG(NA-DJPEG)compression is very common in image tampering.Base on the fact,NA-DJPEG compression detection has attracted great attention in the field of digital forensics.Though some progress and success have been achieved in the detection of NADJPEG compression,most methods are based on grayscale image or use one of color channels,resulting in features of these methods are short of color image's channel correlation and color information.And the design of the method is only suitable for estimating the compression history of the entire image,ignoring the relatively small recompression area.In addition,when the primary compression quality factor(QF1)is greater than the secondary quality factor(QF2),the existing method cannot effectively detect the blocking artifacts generated by the primary JPEG compression.In view of these situations,this paper has conducted in-depth research on it,and the main research contents are as follows:For the problem of missing color image feature information and blocking artifacts detection,this paper proposes a NA-DJPEG compression detection method based on the amplitude-angle feature in spherical coordinates.In the method,three color channels(RGB)are first combined using spherical coordinates to form an amplitude and two angles,so as to make full use of the correlation between color image information and channels.Then,an optimized pixel difference is proposed to detect a given JPEG image.The resulting statistical difference can still well characterize the blocking artifacts of the primary JPEG compression when QF1> QF2.In the end,a new feature set(i.e.,amplitude-angle feature)is obtained on the amplitude and angle,which is used to determine whether a given image has been tampered with.A large number of experiments show that this method is superior to most existing methods.For the small size color image recompression detection problem,a NA-DJPEG compression detection method based on refined Markov model in QDCT domain is proposed in this paper.The reconstructed color image is first mapped to the QDCT domain,and then a new feature selection method is proposed for the transfer probability matrix,so as to improve the Markov model and obtain Markov amplitude feature and Markov angle feature.The refined Markov features can not only capture the intra-block correlation between block QDCT coefficients but also improve computing efficiency in real-time.At last,this method implements NA-DJPEG compression detection on small-sized images(i.e.,64 × 64 pixels).Experiments show that this method is superior to most existing methods.